高速列车自动驾驶非线性广义预测控制方法及应用研究

Application and Research on Nonlinear Generalized Predictive Control Method of High-Speed Train

摘要: 针对高速列车运行过程中具有的强非线性、运行参数时变等问题设计列车自动驾驶的非线性广义预测控制方法。首先采用双线性变换将列车动力学模型离散化为线线模型，将非线性空气阻力描述为未建模动态项，线性模型和未建模动态项组合成集成模型；根据高速列车线路实际运行数据，采用递推辨识算法更新模型中因列车运行环境变化而时变的运行参数，利用极限学习机神经网络估计未建模动态项，建立列车运行过程模型。基于高速列车集成模型，设计以运行速度跟踪为主要控制目标的列车非线性广义预测控制方法。非线性广义预测控制器中，反馈控制器用来控制输出跟踪参考输入，补偿器用来消除未建模动态项对闭环系统的影响，保证列车的速度跟踪精度。仿真结果表明集成模型输出与实际输出的均方根误差为0.1653 km/h，小于线性模型的均方根误差1.454 km/h；并通过与多模型广义预测控制方法的比较显示本文的控制方法跟踪误差更小；所提出算法较BP神经网络方法更加快速。

Abstract: Aiming at the problems of strong nonlinearity and time-varying operating parameters in the operation of high-speed trains, a nonlinear generalized predictive control method for automatic train driving is designed. First, a bilinear transformation is used to discretize the train dynamics model into a linear model, and the nonlinear air resistance is described as an unmodeled dynamic item, and the linear model and unmodeled dynamic item are combined into an integrated model. according to the actual operation of the high-speed train line data, the paper uses the recursive identification algorithm to update the time-varying operating parameters of the model due to changes in the train operating environment, uses the extreme learning machine neural network to estimate the unmodeled dynamic items, and establishes the train operating process model. Based on the high-speed train integration model, a train nonlinear generalized predictive control method with running speed tracking as the main control objective is designed. In the nonlinear generalized predictive controller, the feedback controller is used to control the output to track the reference input, and the compensator is used to eliminate the influence of the unmodeled dynamic term on the closed-loop system and ensure the speed tracking accuracy of the train. The simulation results show that the root mean square error between the output of the integrated model and the actual output is 0.1653 km/h, which is less than the root mean square error of the linear model 1.454 km/h; and the comparison with the multi-model generalized predictive control method shows that by the control method in this paper follows, the error is smaller; the proposed algorithm method is faster than the BP neural network method.